{"id":"https://openalex.org/W7224007632","doi":"https://doi.org/10.48550/arxiv.2610.35697","title":"MiRL: Error-Targeted Reinforcement Learning for Correcting Terminal Agent Misbehaviors","display_name":"MiRL: Error-Targeted Reinforcement Learning for Correcting Terminal Agent Misbehaviors","publication_year":2026,"publication_date":"2026-10-26","ids":{"openalex":"https://openalex.org/W7224007632","doi":"https://doi.org/10.48550/arxiv.2610.35697"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.35697","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.35697","pdf_url":"https://arxiv.org/pdf/2610.35697","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2610.35697","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5007867221","display_name":"Runchu Tian","orcid":"https://orcid.org/0009-0009-7885-1490","observed_orcids":["https://orcid.org/0009-0009-7885-1490"]},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Tian, Runchu","raw_affiliation_strings":["Department of Computer Science University of North Carolina at Chapel Hill"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science University of North Carolina at Chapel Hill","institution_ids":["https://openalex.org/I114027177"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136645574","display_name":"Zaid Khan","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Khan, Zaid","raw_affiliation_strings":["Department of Computer Science University of North Carolina at Chapel Hill"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science University of North Carolina at Chapel Hill","institution_ids":["https://openalex.org/I114027177"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5159797525","display_name":"Zun Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Wang, Zun","raw_affiliation_strings":["Department of Computer Science University of North Carolina at Chapel Hill"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science University of North Carolina at Chapel Hill","institution_ids":["https://openalex.org/I114027177"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5159827682","display_name":"Justin Chih-Yao Chen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chen, Justin Chih-Yao","raw_affiliation_strings":["Department of Computer Science University of North Carolina at Chapel Hill"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science University of North Carolina at Chapel Hill","institution_ids":["https://openalex.org/I114027177"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5125980650","display_name":"Mohit Bansal","orcid":"https://orcid.org/0009-0009-2965-5354","observed_orcids":["https://orcid.org/0009-0009-2965-5354"]},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Bansal, Mohit","raw_affiliation_strings":["Department of Computer Science University of North Carolina at Chapel Hill"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science University of North Carolina at Chapel Hill","institution_ids":["https://openalex.org/I114027177"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I114027177"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.8687000274658203,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.8687000274658203,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10906","display_name":"AI-based Problem Solving and Planning","score":0.022199999541044235,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.022099999710917473,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/long-horizon-tasks","display_name":"long-horizon tasks","score":0.7860000133514404},{"id":"https://openalex.org/keywords/reinforcement-learning-from-human-feedback","display_name":"reinforcement learning from human feedback","score":0.5400000214576721},{"id":"https://openalex.org/keywords/agent-debugging","display_name":"agent debugging","score":0.39500001072883606}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7631000280380249},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6883999705314636},{"id":"https://openalex.org/C168065819","wikidata":"https://www.wikidata.org/wiki/Q845566","display_name":"Debugging","level":2,"score":0.5681999921798706},{"id":"https://openalex.org/C190839683","wikidata":"https://www.wikidata.org/wiki/Q2448197","display_name":"Train","level":2,"score":0.5453000068664551},{"id":"https://openalex.org/C2779664074","wikidata":"https://www.wikidata.org/wiki/Q3518405","display_name":"Terminal (telecommunication)","level":2,"score":0.5281999707221985},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4916999936103821},{"id":"https://openalex.org/C63882131","wikidata":"https://www.wikidata.org/wiki/Q17122954","display_name":"Strengths and weaknesses","level":2,"score":0.4140999913215637},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.34689998626708984},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.3174000084400177},{"id":"https://openalex.org/C2776608160","wikidata":"https://www.wikidata.org/wiki/Q4785462","display_name":"Natural (archaeology)","level":2,"score":0.30149999260902405},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.2770000100135803},{"id":"https://openalex.org/C198104137","wikidata":"https://www.wikidata.org/wiki/Q974688","display_name":"Communication source","level":2,"score":0.25519999861717224}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.35697","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.35697","pdf_url":"https://arxiv.org/pdf/2610.35697","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.35697","doi":"https://doi.org/10.48550/arxiv.2610.35697","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.35697","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2610.35697","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.35697","pdf_url":"https://arxiv.org/pdf/2610.35697","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G572880119","display_name":"CAREER: Semantic Multi-Task Learning for Generalizable and Interpretable Language Generation","funder_award_id":"1846185","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G580734012","display_name":"AI Institute for Engaged Learning","funder_award_id":"2112635","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320333566","display_name":"National Defense Science and Engineering Graduate","ror":null}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7224007632.pdf","grobid_xml":"https://content.openalex.org/works/W7224007632.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7221823890","https://openalex.org/W7212293382","https://openalex.org/W7215831568","https://openalex.org/W7222532804","https://openalex.org/W7204185702","https://openalex.org/W7125695819","https://openalex.org/W7165627739","https://openalex.org/W7223351337","https://openalex.org/W7128611986","https://openalex.org/W7220387536"],"abstract_inverted_index":{"Error-targeted":[0],"learning":[1,36,195],"gives":[2],"a":[3,8,51,91,124],"legible":[4],"way":[5],"to":[6,19,33,37,70,77,93,157,199],"improve":[7],"model:":[9],"identify":[10],"specific":[11],"weaknesses":[12,27],"of":[13,65,90,98,143,187],"the":[14,17,26,88,109,116,130,134,150,153],"model,":[15],"train":[16,78],"model":[18],"fix":[20],"these":[21,119],"weaknesses,":[22],"and":[23,62,72,122,132,175],"check":[24],"whether":[25],"are":[28,57,68],"fixed.":[29],"We":[30,80],"study":[31],"how":[32],"apply":[34],"error-targeted":[35,55,194],"terminal":[38,202],"agents,":[39],"which":[40,85],"solve":[41],"long-horizon":[42,201],"tasks":[43],"specified":[44],"in":[45],"natural":[46],"language":[47],"by":[48],"interacting":[49],"with":[50,111,148],"command":[52],"shell.":[53],"Existing":[54],"methods":[56],"designed":[58],"for":[59,141],"single-turn":[60],"tasks,":[61],"existing":[63],"analyses":[64],"agent":[66],"failures":[67],"used":[69],"evaluate":[71],"debug":[73],"agents":[74],"rather":[75],"than":[76],"them.":[79],"propose":[81],"Misbehavior-driven":[82],"RL":[83],"(MiRL),":[84],"first":[86],"profiles":[87],"errors":[89],"policy":[92,110,151],"find":[94],"generalizable":[95],"misbehaviors:":[96],"kinds":[97],"wrong":[99,144],"steps":[100,117],"that":[101,193],"recur":[102],"across":[103],"many":[104],"environments.":[105],"MiRL":[106,166],"then":[107],"trains":[108],"RL,":[112],"starting":[113],"rollouts":[114],"at":[115],"where":[118],"misbehaviors":[120,139,155],"occur":[121],"rewarding":[123],"rollout":[125],"only":[126],"if":[127],"it":[128,177],"passes":[129],"tests":[131],"avoids":[133],"misbehavior.":[135],"On":[136],"Qwen3.5-4B,":[137],"eight":[138],"account":[140],"52.1%":[142],"steps.":[145],"After":[146],"training":[147],"MiRL,":[149],"repeats":[152],"measured":[154],"43%":[156],"57%":[158],"less":[159],"often":[160],"on":[161,163,169,178],"average":[162],"unseen":[164],"prefixes.":[165],"outperforms":[167],"TMax-4B":[168],"Terminal-Bench":[170,179],"Lite":[171],"(51.8%":[172],"vs.":[173,182],"47.4%)":[174],"matches":[176],"2.1":[180],"(21.3%":[181],"20.6%),":[183],"while":[184],"using":[185],"35%":[186],"its":[188],"rollouts.":[189],"These":[190],"results":[191],"show":[192],"can":[196],"be":[197],"applied":[198],"realistic,":[200],"tasks.":[203]},"counts_by_year":[],"updated_date":"2026-11-06T12:05:46.655746","created_date":"2026-10-28T00:00:00"}